Longitudinal Associations among Depression, substance Abuse, and Crime: A Test of competing Hypotheses for Driving Mechanisms

Abstract

Purpose:

Crime, depression, and substance abuse, often co-occur. This study examined competing models considering each problem domain individually as a driving mechanism for the other problems to better understand etiology and inform prevention efforts. Gender differences were also examined.

Methods:

Data were drawn from the Seattle Social Development Project, a multiethnic and gender-balanced urban panel of 808 participants constituted in 1985. Cross- lagged models examined prospective assessments of early (grades 7 & 8) and late (grades 9–12) adolescent internalizing problems, substance use, and delinquency, as well as measures of depression, substance dependence, and crime at early adulthood (ages 21–24) and later adulthood (ages 27–30).

Results:

Comparisons of nested models by gender showed (a) continuity in internalizing behaviors/depression, substance use/dependence, and delinquency/crime for both women and men; (b) accounting for continuity, depression did not consistently drive other problems for either women or men; (c) among women, both substance abuse and crime appeared to be important driving mechanisms; and (d) among men, crime emerged as the most pertinent driving mechanism.

Conclusion:

Findings suggest that externalizing problems may be more important driving mechanisms for depression than vice versa. Preventing crime and substance abuse may have the important added benefit of reducing early adult depression.

Introduction

Studies have shown that delinquency and crime often co-occur with mental health problems and substance use in both adolescence and adulthood (Huizinga, Loeber, Thomberry, & Cothem, 2000; Snowden, 2001; Abram, Teplin, McClelland, & Dulcan, 2003; Žarkovic Palijan, Mužinić, & Radeljak, 2009). In fact, among juvenile and criminal justice populations, rates of co-occurring substance abuse and mental health problems are much higher than in the general population (Council of State Governments, 2002; Zweig, Schaffer, & Moore, 2004). For example, an estimated 56% of state prison inmates and 64% of local jail inmates are suffering from a mental health problem, while 42% of prison inmates and 49% of jail inmates suffer from both a mental health problem and a substance use problem (James & Glaze, 2006). Female inmates report higher rates of mental health problems (73–75%) than males (55%−63%). In fact, these co-occurrences in justice settings are rather the rule than the exception (Peters, Wexler, & Lurigio, 2015). In the juvenile justice system, upwards of two-thirds of justice-involved youth (70.4%) have at least one behavioral health disorder, including internalizing, externalizing, and substance use disorders (Shufelt & Cocozza, 2006), with females reporting higher rates than males (81.0% vs. 66.8%). Additionally, nearly 61% of justice-involved youth who have a mental health disorder also meet criteria for a substance use disorder (Shufelt & Cocozza, 2006).

While examining rates of co-occurring disorders among justice-involved individuals is informative, not much is learned about the etiology of each problem behavior and its developmental relationship to the others, especially given that some manifestations of both mental health problems and substance use disorders are often criminalized in their own right in the United States (Fisher, Silver, & Wolff”, 2010; Lurigio & Swartz, 2000). This current study seeks to uncover the developmental relationships among delinquency/crime, substance use/abuse, and internalizing behavior/depression across adolescence and adulthood using a longitudinal panel study. Recent studies have suggested that these problems have reciprocal relationships (Merrin, Davis, Berry, D’Amico, & Dumas, 2016; D’Amico, Edelen, Miles, & Morral, 2008; Mason & Windle, 2002) but theoretical arguments are mixed in terms of which of these drives the co-occurrence. Specifically, this paper tests which hypothesis best explains the relationship among these problem behaviors: the crime-driven hypothesis, the mental health-driven hypothesis, or the substance-driven hypothesis. Each hypothesis is detailed below.

Crime-Driven Hypothesis

The crime-driven hypothesis posits that adolescents engaging in antisocial or offending behaviors will experience a cascade of negative problems in other domains of life, such as depressive symptoms, school failure, involvement with antisocial peers, and substance abuse (Capaldi & Stoolmiller, 1999; Patterson & Stoolmiller, 1991). For example, youth who engage in offending behaviors may experience rejection from their peers perhaps due to antisocial behavior and therefore become ftirther involved with like-minded deviant peers engaging in offending and substance using behaviors. As youth become further involved in offending behaviors, they may fail to achieve developmental milestones (e.g., high school graduation) and continue on the path of failure into their adulthood (Moffitt, Caspi, Dickson, Silva, & Stanton, 1996). These accumulated failures will increase the likelihood of continued offending and substance abusing behaviors as well as mental health problems. This study tests to what extent engaging in delinquency and crime drives future experiences of substance use and depression.

Mental Health-Driven Hypothesis

The mental health-driven hypothesis has traditionally been used to explain how substance use and abuse occur as a coping mechanism for mental health problems, such as anxiety and depression. Some researchers have, thus, termed this a self-medication model (Weider & Kaplan, 1969). This hypothesis is popular and has been well-received (Glass, 1990; Khantzian, 1997), and some scholars have expanded the hypothesis to explain not just substance use, but also adolescent delinquent behaviors (Brezina, 2000). Proponents of this hypothesis have argued that youth and adolescents experiencing emotional distress, perhaps as part of normative developmental changes, engage in substance use and delinquent behaviors as a way to redirect or express their negative emotional experience. The empirical support for this hypothesis, however, is, at best, mixed (Damphousse & Kaplan, 1998). It is unclear from most of the cross-sectional or short-term longitudinal designs whether or not mental health problems are indeed driving the increase in substance use and delinquent behaviors. Even less is known about this relationship among adults. Most studies that connect mental health problems and adult offending have been based on samples in prisons or jails (e.g., Shufelt & Cocozza, 2006). This study uses a community sample to examine whether this popular hypothesis is supported.

Substance-Driven Hypothesis

The substance-driven hypothesis has traditionally been used to explain how people who abuse substances have impaired ability to engage in normative activities resulting in the loss of social and personal capital and, in turn, experience mental health problems (Johnson & Kaplan, 1990). For example, those who become dependent on substances might lose employment, disengage from his/her social network, and fail in school. This isolation, in turn, might lead to mental health symptoms like depression. It is also simple to see how substance abuse might lead to delinquency and crime in a similar manner. Studies have shown the connection between substance use and aggression (Pihl & Lemarquand, 1998; Snowden, 2001), suggesting that those with substance use problems may engage in more violent behaviors. Another possibility is that addiction to substances may lead to criminal offending behaviors to support the substance using life style (Anglin & Perrochet, 1998). According to the Bureau of Justice Statistics (2004), about 17–18% of inmates report that they had committed a crime to get money for drugs. In this study, we examine whether substance use drives increases in deviant behaviors and depression.

The current study compares these competing hypotheses using longitudinal panel data spanning several developmental periods. Specifically, this paper tests to what extent these cooccurring problems are driven by delinquency/crime, substance use/dependence, or internalizing behavior/depression. Clarifying the developmental pathways through which these problems occur can provide some empirical findings bearing on current competing theoretical models as well as identify important targets for prevention/treatment efforts. Furthermore, almost all studies using adult samples have examined bivariate relationships between crime and mental health, crime and substance use, and mental health and substance use. This study extends the current knowledge by including all three problem areas to account for the high rates of multivariate comorbidity, in addition to continuity of each problem over time.

Gender Differences

Offending behaviors, substance use, and depression typically occur at different rates by gender, for both adolescents and adults (Bennett, Farrington, & Huesmann, 2005; SAMHSA, 2014; Albert, 2015). Daly and Chesney-Lind (1988) argue that these differences reflect the different social circumstances males and females encounter across development. These varying social circumstances call for theoretically-guided empirical studies that attend to possible differential pathways or processes by which these problems are related. Perhaps for males, earlier offending behaviors drive the occurrence of other problems whereas for females, whose offending rates are much lower than their male counterparts, mental health problems or substance using behaviors lead to offending behaviors. This is particularly important to consider for adolescence, the time in which gender identity becomes solidified (Galambos, Berenbaum, & McHale, 2009) based on different kinds of gender socialization (Bussey & Bandura, 1999). It is also important during the transition to adulthood as social expectations for behaviors vary significantly by gender (Arnett, 2005) and this can affect the ways in which men and women become involved in crime and substance abuse and experience depression. This study examines three competing hypotheses (crime-driven, mental health-driven, and substance-driven) separately by gender to understand whether the developmental relationships among these problem behaviors different across adolescence and early adulthood for males and females.

Methods

Sample

The Seattle Social Development Project (SSDP) is a longitudinal study examining a broad range of behavior and mental health outcomes. The original study population in 1985 included all fifth-grade students in 18 Seattle elementary schools that overrepresented students drawn from high-crime neighborhoods (N = 1,053). From this population, 808 students (77%) consented to participate in the longitudinal study and constituted the SSDP sample. Data for analyses reported here were collected in 7^th, 8^th, 9^th, 10^th, and 12^th grades, and then every 3 years from ages 21 to 33. Retention rates averaged 93% of the still-living sample across these 10 waves (23 participants were deceased by age 33); 806 participants had sufficient data at least at one time point and were included in analyses reported here. Participants were interviewed in person and received monetary compensation. All phases of the study were approved by the Human Subjects Review Committee at the University of Washington.

The sample was gender balanced (51% male) and ethnically diverse (47% White/European American, 26% Black/African American, 22% Asian/Asian American, 5% Native American). As children, 52% of participants were eligible for the federal school lunch/breakfast program at some point in the fifth, sixth, or seventh grade.

Measures

It is important to note that measures changed slightly over the study period to reflect developmental changes in behavior (e.g., participants were asked about using illegal checks in adulthood, but not in adolescence). Thus, our language also differs slightly regarding constructs measured in adolescence versus adulthood (e.g., delinquency versus crime and internalizing behaviors versus depression), similar to studies that show continuity between conduct disorder in adolescence and antisocial personality in adulthood (Loeber, Burke, Lahey, 2002; Simonoff et al., 2004). While the items measured might differ somewhat, the study sought to assess the same general construct at each developmental phase.

Delinquency in adolescence and crime in adulthood.

We used a weighted measure of adolescent delinquency that took into account both severity and frequency: a count of the number of self-reported delinquent acts committed in the past year weighted by severity (one point for minor acts, two points for moderate acts, and three points for serious acts). Examples of minor acts of delinquency included stealing something worth less than $5, picking a fight, and drawing graffiti. Examples of moderate delinquency included damaging or destroying property, hitting parents, and stealing something worth less than $50. Examples of serious delinquency included breaking into a building, using a weapon or force to get something, and stealing something worth more than $50. Each youth was assigned a value at each survey year based on this weighted measure. To address skewness due to outliers, data were truncated at the 98th percentile, so that the top 2% of youth were assigned the value at the 98th percentile. The weighted frequencies were then averaged across 7th and 8th grades (α = .88) for early adolescence and then across 9th, 10th, and 12th grades (α = .86) for late adolescence.

Similar to the adolescent delinquency measures, we used a weighted measure of adult crime consistent with prior research (Blumstein, 1974, Kim, Gilman, Hill, & Hawkins, 2016, Gilman, Hill, Kim, Nevell, Hawkins, & Farrington, 2014). Examples of minor crime include stealing something worth less than $50, using illegal checks, and using another person’s credit card. Examples of moderate crime include damaging or destroying property, hitting others with the idea of hurting them, and stealing something worth more than $50. Examples of serious crime include beating someone badly enough to need medical treatment, using a weapon or force to get something, and drug selling. The severity-weighted frequencies were averaged across ages 21 and 24 (α = .69)¹ for early adulthood and then across ages 27, 30, and 33 (α = .72) for later adulthood, and data were truncated at the 98th percentile.

Internalizing problems in adolescence and depression in adulthood.

Internalizing problems were measured with items adopted from the Child Behavior Checklist (Achenbach & Edelbrock, 1983), summing across 10 self-reported items assessing withdrawn and anxious behaviors (e.g., likes to be alone, shy, worries a lot) (ranging from 0 to 20). The frequencies were then averaged across 7th and 8th grades (α = .77) for early adolescence and then across 9th, 10th, and 12th grades (α = .81) for late adolescence.

Depressive symptoms were assessed by self-report at ages 21, 24, 27, 30, and 33 with a modified version of the Diagnostic Interview Schedule (DIS) (McGee et al, 1990), which measured criteria specified by the DSM-IV (American Psychiatric Association, 1994). The DIS has been used frequently in studies of psychiatric disorders among adults drawn from the general population and has been demonstrated to be valid and reliable (Newman et al., 1996; Reinherz et al., 2000). Measures of depressive symptoms summed criteria met for major depressive episode (e.g., anhedonia, weight changes, sleep problems) in the past year at each adult wave, ranging from 0 to 9. These symptom counts were averaged across ages 21 and 24 (α= .94) for early adulthood and then across ages 27, 30, and 33 for later adulthood (α = .96).

Substance use in adolescence and abuse in adulthood.

The frequency of alcohol use and 9 different illicit substances (e.g., marijuana, crack, cocaine, amphetamine, tranquilizers) in the past year were summed to create a substance use measure. The frequencies were then averaged across 7th and 8th grades (range = 0–41; α = .84) for early adolescence and then across 9th, 10th, and 12th grades (range = 0–151; α = .74) for late adolescence. To address skewness due to outliers, data were truncated at the 95th percentile, so that the top 5% of youth were assigned the value at the 95th percentile.

In adulthood, substance use was measured with a count of the number of criteria met for alcohol and illicit drug abuse and dependence (e.g., neglect of responsibilities, increased tolerance, withdrawal symptoms) in the past year at each adult wave. Scores ranged from 0 to 22 (11 criteria each for alcohol and illicit drugs, except at age 21 where one criterion - drug withdrawal symptoms - was not measured). Again, these symptom counts were averaged across ages 21–24 (α = .91) for early adulthood and then across ages 27, 30, and 33 (α = .94) for later adulthood.

To help control for the effect of extreme values, all measures were natural-log transformed for analyses. Correlations among measures for females and males are shown in Table 1. For descriptive purposes, the mean untransformed measures are also shown.

Table 1.

Correlations for longitudinal associations.

		1	2	3	4	5	6	7	8	9	10	11	12	Males
														Mean	SD
1	Internali zing behavio r early adolesce nce		0.50***	0.01	0.12*	0.03	0.09	0.12*	0.11	0.01	0.06	0.09	0.002	5.65	2.67
2	Internali zing behavio r late adolesce nce	0.54***		0.02	0.10*	0.03	0.04	0.22***	0.125	0.06	0.07	0.03	0.06	3.33	1.88
3	Substan ce use early adolesce nce	0.09	0.13**		0.38***	0.63***	0.29***	0.1	0.08	0.16 **	0.15 **	0.05	0.14**	5.79	11.1 1
4	Substan ce use late adolesce nce	0.03	0.10*	0.44 ***		0.35***	0.51***	0.11*	0.15**	0.44 ***	0.28 ***	0.18 ***	0.23***	30.22	45.6 3
5	Delinqu ency early adolesce nce	0.02	0.02	0.39 ***	0.37***		0.46***	0.11*	0.15**	0.18 ***	0.15 **	0.12 *	0.22***	16.20	38.4 4
6	Delinqu ency late adolesce nce	0.04	0.001	0.26 ***	0.40***	0.54***		0.11*	0.07	0.17 ***	0.12 *	0.24 ***	0.19***	53.23	132. 06
7	Depressi on early adultho od	0.17***	0.21***	0.20 ***	0.14**	0.10*	0.08		0.35***	0.32 ***	0.14 **	0.25 ***	0.07	1.90	2.35
8	Depressi on later adultho od	0.10*	0.24***	0.12 *	0.12*	0.07	0.02	0.39***		0.23 ***	0.38 ***	0.12 *	0.14**	1.04	2.03
9	Substan ce abuse early adultho od	0.02	0.06	0.23 ***	0.17***	0.20***	0.06	0.23***	0.28***		0.54 ***	0.31 ***	0.28***	2.60	2.85
10	Substan ce abuse later adultho od	0.005	0.09	0.21 ***	0.20***	0.19***	0.096	0.22***	0.37***	0.57 ***		0.22 ***	0.40***	1.66	2.46
11	Crime early adultho od	0.02	0.03	0.04	0.15**	0.03	0.03	0.02	0.02	0.29 ***	0.11 *		0.20**	18.86	58.5 6
12	Crime later adultho od	0.004	0.06	0.14 **	0.08	0.29***	0.06	0.09	0.06	0.30 ***	0.34 ***	0.31 ***		7.82	25.2 6
Females	Mean	6.18	3.91	5.86	17.88	9.07	15.07	2.44	1.54	1.20	.82	3.66	1.69
	SD	2.63	2.05	11.0 7	33.04	25.26	63.39	2.69	2.38	1.93	2.03	23.0 9	10.32

Analyses

In the current analyses, a cross-lagged panel model was used to explore the relationship among observed measures of delinquency/crime, internalizing behavior/depression, and substance use/dependence across four developmental time periods: early adolescence (grades 7 & 8); late adolescence (grades 9–12); early adulthood (ages 21–24); and later adulthood (ages 27– 33). Cross-lagged panel models examine the relationship among phenomena that are repeatedly measured overtime (Selig & Little, 2012) and they can be particularly useful when determining the direction of the association among two or more co-occurring events. More specifically, the cross-lagged model used in the current analyses allows the assessment of the stability of delinquency/crime, internalizing behavior/depression, and substance abuse/dependence overtime, and in addition, the time-ordered effect of each phenomenon on the others, controlling for previous levels (Norona & Baker, 2014). As described below, we estimated five cross-lagged models, each testing a different hypothesis, and then compared model fit to determine which best fit the data. Models were estimated separately for males and female.

Analyses² were conducted in MPlus version 6.11 (Muthen & Muthen, 1998–2011) using full-information maximum likelihood estimates with incomplete data. The comparative fit index . (CFI; Bentler, 1990) and root mean square error of approximation (RMSEA; Browne & Cudeck, 1993) are reported for overall model fit.

Results

Models consistent with the competing theoretical perspectives described earlier were tested sequentially, as shown in Figure 1. Panel A shows an autoregressive or continuity model where each outcome is regressed on itself over time. This represents the initial comparison model against which other models were tested for improvement in fit. The model accounts for stability in each outcome from one time point to the next, as well as more distal stability associations spanning from 1 to 21 years. Concurrent associations between different outcome domains are also included in this model. Panel B, the depression-driven model, builds on the autoregressive model by adding cross-lagged paths from internalizing problems and depression symptoms to substance use and crime outcomes at the next time point. Similarly, Panels C and D add cross-lagged paths to the autoregressive model consistent with a sub stance-driven model and a crime-driven model, respectively.

Figure 1.

Models of Longitudinal Associations Panel A: Autoregressive model assuming no cross-lagged effects. Panel B: Depression-driven model assuming cross-lagged effects of internalizing problems/depression on substance use/abuse and delinquency/crime. Panel C: Substance-driven model assuming cross-lagged effects of substance use/abuse on depression and delinquency/crime. Panel D: Crime-driven model assuming cross-lagged effects of delinquency/crime on internalizing problems/depression and substance use/abuse.

As reviewed earlier, research indicates gender differences in each of the adult outcome domains examined here. As a first step, we tested the equivalence of covariances among model constructs between females and males by comparing the fit of an unconstrained multiple-group model (in which covariances were allowed to vary freely) with a constrained model (in which covariances were constrained to be equal for females and males). The constrained model showed a significant decrement in fit (Δχ2(66) = 141.53, p < .001), and modification indices suggested several possible gender differences. Subsequent analyses were conducted separately for females and males.

Tables 2 and ³show a systematic comparison among females and males, respectively, of the models depicted in Figure 1.

Table 2.

Comparison of nested models for females

Tested model	χ2(df)	P	CFI	RMSEA	Comparison model	Δχ2(∆df)	p of Δ
Autoregressive	124.63(36)	.000	.916	.079
Depression-driven	117.89(30)	.000	.916	.086	Autoregressive	−6.74(6)	.346
Substance-driven	66.22(30)	.000	.965	.055	Autoregressive	−58.41(6)	.000
Crime-driven	58.76(30)	.001	.973	.049	Autoregressive	−65.87(6)	.000
Sub. + crime-driven	21.58(24)	.605	1.000	.000	Substance-driven	−44.64(6)	.000
					Crime-driven	−37.18(6)	.000

Note. N = 395 for all models.

Longitudinal Associations for Females

In Table 2, fit indices for females are shown. The continuity model (Figure 1, Panel A) including all autoregressive relationships as well as concurrent relationships across domains resulted in moderate fit with the data. The depression-driven model (Figure 1, Panel B), adding 8 cross-lagged paths to the continuity model did not significantly improve model fit (indicated by a non-significant Δχ2). Both the substance-driven and the crime-driven models (Figure 1, Panels C and D), also compared with the autoregressive model, however, did show significant improvements in fit. The final two rows of Table 2 then tested a “dual process” model combining all 16 cross-lagged paths from both the sub stance-driven and crime-driven models and compared it, respectively, with the two component models. The combined substance- and crime-driven dual process model was found to be the best fitting model overall for females, fitting significantly better than either the substance-driven or crime-driven model alone.

The combined substance- and crime-driven model for females is shown in Figure 2; for clarity, only significant paths are shown. Autoregressive paths indicated a great deal of continuity for each outcome over time with one exception: Internalizing problems in late adolescence did not significantly predict depression in early adulthood. Several significant autoregressive paths spanned 9 or more years (up to 21 years in the case of delinquency and crime), suggesting distal relationships in each domain of outcomes. Delinquency during early adolescence significantly predicted crime during later adulthood and internalizing problems and substance use during later adolescence significantly predicted depression and substance dependence, respectively, during later adulthood. Many significant concurrent associations across outcome domains were found as well, particularly between substance use/dependence and delinquency/crime outcomes, which were significantly correlated with each other at all time points. Internalizing problems/depression was significantly correlated with substance use/dependence at all time points except for early adolescence. During early adulthood, depression and crime were also significantly correlated.

Figure 2.

Substance- and Crime-driven model for females For clarity, only significant paths are shown. Standardized path coefficients are included for significant autoregressive and cross-lagged paths. * p < .05, **p < .01, *** p < .001.

Controlling for continuity and concurrent relationships, several cross-lagged effects were indicated. Specifically, substance use in early adolescence predicted involvement in delinquent behaviors in late adolescence, which, in turn, predicted depression in early adulthood. Delinquency in early adolescence predicted substance use in late adolescence and substance dependence in early adulthood predicted depression in later adulthood.

Longitudinal Associations for Males

Table ³ provides model fit indices for males. Fit indices suggest that a model including all autoregressive relationships, as well as concurrent relationships across domains (Figure 1, Panel A), resulted in moderate fit with the data. The depression-driven (Figure 1, Panel B), substance-driven (Figure 1, Panel C), and crime-driven (Figure 1, Panel D) models, compared with the autoregressive model, showed significant improvements in fit. The final two rows indicate that the combined substance- and crime-driven dual process model significantly improved model fit compared to the substance-driven model but not compared to the crimedriven model. Together, these model comparison indices suggest that, for males, the crimedriven model fit the data the best.

The crime-driven model for males is shown in Figure 3, with only significant paths shown. As observed among females, autoregressive paths for males also showed a strong continuity for delinquency/crime, internalizing behavior/depression, and substance use/dependence. Significant associations were found across each adjacent measurement point. Substance use/dependence and delinquency/crime were significantly correlated with each other at each time point across all developmental periods. Depression was significantly correlated with substance dependence and crime during adulthood. Different from females, only one significant autoregressive path indicated distal relationships for males where delinquency in early adolescence predicted crime in early adulthood.

Figure 3.

Crime-driven model for males For clarity, only significant paths are shown. Standardized path coefficients are included for significant autoregressive and cross-lagged paths. * p < .05, ** p < .01, *** p < .001.

Controlling for continuity and concurrent relationships, several cross-lagged effects were indicated for males as well. Delinquency in early adolescence significantly predicted substance use in late adolescence while delinquency in late adolescence led to depression and substance dependence in early adulthood. Furthermore, crime in early adulthood predicted substance dependence in later adulthood.

Discussion

This paper tested three competing hypotheses to examine the longitudinal relationship among delinquency/crime, internalizing behaviors/depression, and substance use/dependence from adolescence to adulthood. The crime-driven hypothesis suggests that offending behaviors, such as delinquency and crime, lead to more general life instability and failure and, in turn, depression and substance use. The depression-driven hypothesis suggests that externalizing behaviors, such as delinquency/crime and substance use/dependence, could occur either as a visible expression of emotional distress or a coping strategy to depression. Finally, the substance use-driven hypothesis suggests that reliance on substances impairs functional ability and may isolate individuals from social and human capital leading to depression, or leads to criminal involvement to further access substances. To ensure temporal ordering of these relationships, this paper used longitudinal data spanning over 20 years across 10 waves, while accounting for continuity of each problem over time.

Consistent with previous research on the continuity of mental, emotional, and behavioral health problems overtime (Coie et al., 1993; O’Connell, Boat, & Warner, 2005), the autoregressive model suggested strong continuity of each of these problems from adolescence to adulthood for both males and females. This is further evidence that prevention of these problems early on could have a lifelong impact, and that prevention or treatment programs at any developmental period could potentially be an opportunity to break the continuity of these problems. Moreover, the findings in this study indicated that substance use/dependence was significantly correlated with delinquency/crime at each time point for both males and females, confirming the high likelihood of concurrent comorbidity (e.g., Loeber et al., 2000) between these two problems. Evidence-based programs that target common risk and protective factors (e.g., attitudes towards antisocial behaviors, refusal skills) may be particularly important to prevent the early occurrences of these problems.

We found that different hypotheses described the longitudinal pathways involving crime for females and males. For females, the combined substance use and crime-driven hypothesis fit the data best. For the most part, engaging in substance use and offending behaviors led to more depressive symptoms. Contrary to the popular hypothesis for women and girls that internalizing problems or mental health issues lead to externalizing or offending behaviors, our data suggested that the opposite was true. In fact, externalizing behaviors (i.e., crime/delinquency and substance use) were driving the increased symptoms of internalizing behaviors. This could stem from the possibility that females who engage in offending behaviors, which is less common among females and perhaps less accepted, become socially isolated and thus become depressed. Because the rates of offending behaviors among females are much lower than males, crime literature has focused less on females (Daly & Chesney-Lind, 1998; Chesney-Lind, 2006). Similarly, because depression rates are higher among females than males, mental health research and practice have focused more on females (Albert, 2015). Our findings suggest, however, that focusing on preventing early occurrences of externalizing behaviors among females could potentially reduce and prevent the occurrences of depression.

For males, the crime-driven hypothesis was supported by the data. Different from their female counterparts for whom both delinquency/crime and substance use/abuse were important drivers, engaging in offending behaviors for males best predicted both later depression and substance using behaviors. Again, our data did not support the mental health-driven hypothesis that people with mental ailments “self-medicate.” Rather, earlier externalizing behaviors, specifically offending behaviors, led to increased substance use and depression later. Researchers agree that male youth engage in delinquent behaviors at much higher rates than their female counterpart (Bennett et al., 2005). Given the long-term consequences of engaging in delinquent behaviors, including potentially leading to substance dependence and depression later in life, it is prudent not to dismiss male delinquency as “boys being boys.” Prevention and intervention for males, therefore, may be optimized if universal in nature reaching all male youth in schools rather than few selective youth exhibiting serious behavioral problems.

The study has limitations. The study was based on a nonrepresentative sample originally drawn from public schools in a single western metropolis. It is also worth stating explicitly that the significant pathways found in the analyses reported do not entirely discount the possibility of other important etiological relationships. However, the reported findings should contribute to our broad understanding of what causes what, to be explored further in different models with different samples. All measures were based on self-reported data. Studies have shown that selfreport of delinquency/crime are fairly reliable when compared to official records (e.g., Gilman et al., 2014), and the long history of protecting confidential data collected from the SSDP sample should buttress confidence in honest self-reports. We did not control for race, socioeconomic status, or criminal justice involvement, which could affect the longitudinal pathways; however, the focus of the study was to provide empirical evidence for the three competing hypotheses and thus sought to model the temporal ordering for these problem areas in relatively parsimonious models. Future studies should examine individual and contextual characteristics (i.e., race, socioeconomic status, trauma) that might moderate the relationships found in this study. Also, differential longitudinal relations might be found for those involved in public systems (e.g., child welfare, juvenile justice, criminal justice). Mechanisms by which these hypotheses are supported should be explored as well.

The strength of this study is the use of longitudinal data to examine the relationship among delinquency/crime, substance use/dependence, and internalizing behavior/depression across the lifespan, covering multiple developmental periods. This allowed us to model temporal ordering and provided an optimal opportunity to test these hypotheses. Furthermore, our sample included enough females to examine the development of these problem areas separately by gender. Daly and Chesney-Lind (1988) urged all criminologists to address gender differences by examining theoretical models separately for boys and girls. Finally, the findings contribute to empirical tests of these competing hypotheses that have important implications for prevention and treatment of these problem areas.

Table 3.

Comparison of nested models for males

Tested model	χ2(df)	P	CFI	RMSEA	Comparison model	∆χ2(∆df)	p of Δ
Autoregressive	108.60(36)	.000	.938	.070
Depres s ion-driven	95.68(30)	.000	.944	.073	Autoregressive	−12.92(6)	.044
Substance-driven	79.46(30)	.000	.958	.063	Autoregressive	−29.14(6)	.000
Crime-driven	40.12(30)	.103	.991	.029	Autoregressive	−68.48(6)	.000
Sub. + crime-driven	31.97(24)	.128	.993	.028	Substance-driven	−47.49(6)	.000
					Crime-driven	−8.15(6)	.227

Note. N = 411 for all models.